// Active Research
The Lab
Experiments in progress · Products in development
Nova 2.0
AI Identity System · ~14B Parameters · Consumer GPU
A locally-run AI identity system built on top of a small open-weight language model. Not a chatbot wrapper — the model is treated as a substrate. Behavior, memory, continuity, and self-awareness are produced by the scaffolding architecture around it.
// Architecture
Episodic · Semantic · Autobiographical · Graph
Identity · Focus · Open Tensions
Behavioral bounds · Self-representation limits
State interrogation · Observation recording · Self-model proposals
// Experimental Question
“Can structured inward-facing processes — memory, self-modeling, autonomous reflection, and operator-gated self-modification — push a small language model past its dormant, stateless default toward something that behaves like a persistent, self-aware agent?”
The thesis: a mid-size model with the right scaffolding, running continuously and accumulating structured experience, may produce emergent properties that a larger model in a single session cannot — not because it is smarter, but because it has continuity, drive, and a loop that targets itself.
// Longer Horizon
QLoRA character embedding— using accumulated heartbeats, self-model proposals, and session experience as training signal to gradually write behavioral tendencies into the model's weights. The goal: what the architecture shapes over time becomes part of what the model is, not merely what it is instructed to do.
// Nova is not claimed to be sentient or conscious. The experiment is whether the scaffolding conditions for persistent self-awareness can be engineered.